ON CONNECTIVITY IN MATROIDS AND GRAPHS
Transcript of ON CONNECTIVITY IN MATROIDS AND GRAPHS
TRANSACTIONS of theAMERICAN MATHEMATICAL SOCIETYVolume 265, Number I, May 1981
ON CONNECTIVITY IN MATROIDS AND GRAPHS
BY
JAMES G. OXLEY
Abstract. In this paper we derive several results for connected matroids and use
these to obtain new results for 2-connected graphs. In particular, we generalize
work of Murty and Seymour on the number of two-element cocircuits in a
minimally connected matroid, and work of Dirac, Plummer and Mader on the
number of vertices of degree two in a minimally 2-connected graph. We also solve
a problem of Murty by giving a straightforward but useful characterization of
minimally connected matroids. The final part of the paper gives a matroid
generalization of Dirac and Plummer's result that every minimally 2-connected
graph is 3-colourable.
1. Introduction. The structure of minimally 2-connected graphs was determined
independently by Dirac [5] and Plummer [16]. Their work led Murty [10] to
examine minimally 2-connected matroids and some of the latter's results were
generalized by Seymour [17], [18]. In §2 of this paper, we strengthen one such result
of Seymour by showing that if C is a circuit in a 2-connected matroid M and for all
x in C, the restriction M \ x is not 2-connected, then provided |£(A/)| > 4, M has
at least two disjoint cocircuits of size two contained in C. Several corollaries of this
theorem are proved and the theorem is also used to derive the corresponding result
for a 2-connected graph G, the conclusion in this case being that the circuit C
meets at least two nonadjacent vertices of G of degree two. This result, which
extends a result of Dirac [5] and Plummer [16] for minimally 2-connected graphs, is
a strengthening in the case n = 2 of a result of Mader [8] for «-connected graphs. It
has a number of corollaries including a new lower bound on the number of vertices
of degree two in a minimally 2-connected graph. Some similar results for minimally
«-connected graphs and matroids are also obtained for n > 3.
In [10], Murty asks for a characterization of minimally 2-connected matroids. In
§3, we give such a result, showing that unless every element of a minimally
2-connected matroid M is in a cocircuit of size two, M can be obtained from two
minimally 2-connected matroids on fewer elements by a join operation which is
closely related to series connection. This characterization, which is not difficult to
prove, is used to give short proofs of the main results of [10].
Dirac [5] and Plummer [16] have shown that a minimally 2-connected graph is
3-colourable and Dirac's argument [5, p. 215] can be extended to show that a
minimally «-connected graph is (« + l)-colourable (see, for example, [1, Corollary
4.7]). In §4, by generalizing this argument, we establish the corresponding result for
minimally «-connected matroids when « = 2 or 3.
Received by the editors October 19, 1979.
1980 Mathematics Subject Classification. Primary 05B35, 05C40, 05C15; Secondary 05C35.
© 1981 American Mathematical Society
0002-9947/81/0000-0202/$04.00
47
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
48 J. G. OXLEY
The terminology used here for matroids and graphs will in general follow Welsh
[23] and Bondy and Murty [2] respectively. Most of the known graph connectivity
results to which we shall refer can be found in Bollobás [1, Chapter 1]. The ground
set of a matroid M will be denoted by E(M) and, if T G E(M), we denote the
rank of T by rk T. We shall write rk M for rk(E(M)) and cork M for the rank of
the dual matroid M* of M. The restriction of M to E(M) \ T will sometimes be
denoted by M \ T or, if T = {xx, x2, . . . , xm), by M \ xx, x2, . . . , xm. Likewise,
the contraction of M to E(M) \ T will sometimes be written as M/ T or
M/xx, x2, . . . , xm. A cocircuit of M having k elements will be called a k-cocircuit,
and a ^-element circuit, a k-circuit. If x,y G E(M) and {x,y} is a 2-cocircuit, we
say that x and y are in series; if instead {x,y} is a 2-circuit, then x and_y are in
parallel. A series class of M is a maximal subset A of E(M) such that if a and A are
distinct elements of A, then a and A are in series. Parallel classes are defined
analogously. We call a series or parallel class nontrivial if it contains at least two
elements.
Familiarity will be assumed with the concept of «-connection for graphs as
defined, for example, in [2, p. 42]. We now recall the definition of «-connection for
matroids [20]. If Ac is a positive integer, the matroid M is k-separated if there is a
subset Tof E(M) such that \T\ > k, \E(M) \T\> k and
rk T + rk(E(M) \ T) - rk M = k - 1.
If there is a least positive integer j such that M is y'-separated, it is called the
connectivity k(M) of M. If there is no such integer, we say that k(M) = oo. The
matroid M is said to be n-connected for any positive integer « such that n < k(M).
It is routine to show [20, (12)] that
k(M) = k(M*). (1.1)
The familiar notion of connectivity or nonseparability of matroids is related to
«-connectedness as follows.
(1.2) A matroid is connected if and only if it is 2-connected.
A matroid or graph H is minimally n-connected if H is «-connected and, for all
elements e of E(H), H \ e is not «-connected.
The notions of «-connectedness of a graph G and «-connectedness of its cycle
matroid M(G) do not, in general, coincide (see [6], [21], [23]). However, [21, pp.
1-2]
(1.3) // G has no loops and at least three vertices, then G is 2-connected if and only
if M(G) is 2-connected; and
(1.4) // G is simple and has at least four vertices, then G is 3-connected if and only
if M(G) is 3-connected.
If G is a 2-connected graph without loops, then clearly the set of all edges
incident with a vertex of G is a cocircuit of M(G). Such a cocircuit will be called a
vertex cocircuit of M(G).
2. Connectivity results for matroids and graphs. The following basic lemmas will
be used frequently throughout this paper.
(2.1) Lemma [20, (6.5)]. If M is a 2-connected matroid and e G E(M), then either
M \ e or M/e is 2-connected.
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
CONNECTIVITY IN MATROIDS AND GRAPHS 49
(2.2) Lemma [15, Lemma 2.1]. Let M be a matroid and n be an integer exceeding
one. Suppose that e is an element of M for which M/e is n-connected but M is not.
Then either e is a loop of M or M has a cocircuit containing e and having fewer than
n elements.
In this section we shall give two extensions of a result of Seymour [18, (2.3)]. The
next lemma is the first of these.
(2.3) Lemma. Let M be a 2-connected matroid having at least two elements and
{xx, x2, . . . , xm) be a circuit of M such that M \ x¡ is not 2-connected for all i in
(1, 2, . . . , m — 1}. Then {xx, x2, . . . , xm_j) contains a 2-cocircuit of M.
Proof. We argue by induction on m. If m = 1 or 2, the result is vacuously true,
so suppose m > 2. As M \ xx is not 2-connected, it follows by Lemma 2.1 that
M/xx is 2-connected. Moreover, {x2, x3, . . . , xm] is a circuit of M/xx. Thus, if for
all i in (2, 3, . . . , m - 1}, M/xx \ x¡ is not 2-connected, then, by the induction
assumption, {x2, x3, . . . , xm_,} contains a 2-cocircuit of M/xx, and hence, as
required, {x,, x2, x3, . . . , xm_,} contains a 2-cocircuit of M. We may therefore
suppose that for some i in (2, 3, . . . , m — 1}, M/xx \ x, is 2-connected. Thus, as
M \ x, is not 2-connected, by Lemma 2.2, x, is a loop or a coloop of M \ x, and so
{x,, x,} is a 2-cocircuit of M and the required result holds. □
In the notation of Lemma 2.3, Seymour [18, (2.3)] proved that {x,, x2, . . . , xm)
contains a 2-cocircuit provided M satisfies the additional condition that M \ xm is
not 2-connected. Earlier he [17, (2.2)], Murty [10, Lemma 3.1] and White [24,
Lemma 4.6] had independently shown that a minimally 2-connected matroid with
at least two elements has a 2-cocircuit.
The next theorem, an alternative strengthening of Seymour's result, underlies
most of the remaining results in this section.
(2.4) Theorem. Let C be a circuit of a 2-connected matroid M such that for all x in
C, M \ x is not 2-connected. Then either M is the circuit C or C contains at least two
distinct nontrivial series classes of M.
Proof. We argue by induction on |C|. If \C\ = 1, then the result is trivially true.
Now suppose that \C\ > 2. Then, by Lemma 2.3, M has a 2-cocircuit (x,y)
meeting C. If every element of C is in series with x, then it follows, since M is
2-connected, that C = E(M); for, if there is an element e in E(M) \ C, then M has
a circuit C containing e and meeting C. But, as a circuit and a cocircuit of M
cannot have exactly one common element, we obtain the contradiction that
C ^ C. It follows that we may suppose that C contains an element z which is not
in series with x. Consider M/z. By Lemma 2.1, this matroid is 2-connected.
Moreover, C \ z is a circuit of M/z. Now, if a is an element of C \ z, then
consider M/z \ a. If this is 2-connected, then, as M \ a is not 2-connected, it
follows from Lemma 2.2 that {a, z) is a 2-cocircuit of M. Evidently {a, z) G C,
and as z is not in series with x, the required result holds. Thus we may assume that
M/z \ a is not 2-connected for all a in C \ z. The required result now follows on
applying the induction assumption. □
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
50 J. G. OXLEY
The preceding result is best-possible for if we take the cycle matroid of the plane
graph G shown in Figure 1, the cycle C bounding the exterior face of G contains
only two distinct nontrivial series classes. It is not difficult to extend this to give
examples of matroids in which | C | is arbitrarily large and yet C still contains only
two nontrivial series classes.
The next result, which generalizes a graph-theoretic result of Dirac [5, Proposi-
tion 3] and Plummer [16, Corollary lb], is an immediate consequence of the
preceding theorem.
(2.5) Corollary. // a 2-connected matroid M has a 3-circuit T such that for all x
in T, the matroid M \ x is not 2-connected, then M is the 3-circuit T.
(2.6) Corollary. Let M be a 2-connected matroid other than a single circuit.
Suppose that A G E(M) such that for all a in A, M \ a is not 2-connected. Then
either A is independent or A contains at least \A\ — rk A + 1 nontrivial series classes
of M.
Proof. Suppose that A is dependent and let A ' be the union of all circuits of M
contained in A. We shall prove that A' contains at least \A'\ — rk A' + 1 nontrivial
series classes of M. The required result will then follow since \A\ — rkA = \A'\ —
rk A '. As a circuit and a cocircuit of a matroid cannot have exactly one common
element, if C* is a 2-cocircuit of M meeting A', then C* G A', and moreover, C* is
a 2-cocircuit of M\A'. Now let X be the set of elements of A' which are contained
in a 2-cocircuit of M. Then, by Theorem 2.4, X meets every circuit of M\A'. Since
the cobases of a matroid are the minimal sets meeting every circuit, it follows that
X contains a cobase B* of M\A'. Choose an element A of B* and let Cb be the
fundamental circuit of A in M\A' with respect to A' \ B*. By the choice of B*, M
has a 2-cocircuit, say {A, c), containing A. Moreover, by Theorem 2.4, Cb contains
another 2-cocircuit {d, e) which is contained in a different series class from (A, c}.
Now let B* = {x,, x2, . . . , xm) where x, = A, and let {x,, yx),
{x2,y2}, ■ ■ ■ , {xm,ym} be 2-cocircuits of M where yx = c. We complete the proof
by showing that {xx,yx), {x2,y2), . . . , {xm,ym), {d, e) are contained in different
series classes of M. Since B* is a cobase of M\A', it contains at most one element
of each series class of M and therefore {x,,y,}, {x2,y2), . . . , {xm,ym} are
contained in distinct nontrivial series classes of M. It remains to show that none of
these classes contains {d, e). But, if for some /, {d, e, x,, yj) is contained in a series
class of M, then this series class is contained in Cb. It follows that 1=1, hence
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
CONNECTIVITY IN MATROIDS AND GRAPHS 51
{x¡,y¡} = i0' c) anc* we ^ave a contradiction to the fact that {b, c) and {d, e) are
contained in different series classes of M. fj
The next result is an immediate consequence of the preceding corollary, slightly
strengthens [14, Proposition 4.13].
(2.7) Corollary. Let M be a minimally 2-connected matroid which is not a
circuit. Then M has at least cork M + l nontrivial series classes and therefore has at
least cork M + 1 pairwise disjoint 2-cocircuits.
We now come to apply Theorem 2.4 to graphs. First, however, we note the
following result of Mader.
(2.8) Theorem [8, Satz 1]. Let C be a cycle of a simple n-connected graph G such
that for all edges x of C, G \ x is not n-connected. Then C meets a vertex which has
degree n in G.
In the case n = 2, we can strengthen this result as follows.
(2.9) Theorem. Let G be a 2-connected graph without loops and C be a cycle of G
such that for all edges x of C, the graph G \ x is not 2-connected. Then either G is the
cycle C or C meets two vertices of G of degree two which are separated on C by
vertices of degree greater than two.
Proof. As G is 2-connected, | V(G)\ > 3 and therefore, by (1.3), M(G) is
2-connected. The required result will follow if we can show that either G is a cycle
or C contains at least two nontrivial series classes of M(G) each of which contains
a vertex 2-cocircuit of G. We shall show this by induction on \C\ noting that the
result holds for \C\ < 3 (see Corollary 2.5). Now assume that G is not a cycle.
Then, by Theorem 2.4, C contains at least two nontrivial series classes, Ax and A2
say. We may assume that Ax does not contain a vertex 2-cocircuit of M(G). Let
{a, A} be a 2-cocircuit contained in Ax. Then G \ {a, b) has precisely two compo-
nents, Hx and H2 say. Let a = (ax, a2) and A = (A,, A^. Then clearly ax, a2, A, and
A2 are distinct. For / = 1,2, assume that a,, A, G //,, and form G, as follows. Let
V(Gj) = V(Hj) u {/} and E(Gj) = E(Hj) u {(a„ 0, (A,., /)}. Then clearly G, and G2
are both 2-connected. Moreover, neither G, nor G2 is a cycle as otherwise Ax
contains a vertex 2-cocircuit of G. Now for / = 1,2, let
C, = (C n E(Hj)) u {(a„ /), (A,., ,)}-
Then evidently C, is a cycle of G,. In addition, it is not difficult to check that for all
x in C¡, the graph G, \ x is not 2-connected. Then, as |C,| < |C|, it follows by the
induction assumption that C, contains at least two nontrivial series classes AiX and
Ai2 of M(Gj) each of which contains a vertex 2-cocircuit of G,. Assume without loss
of generality that {(a¡, i), (A,, /')} G Ai2. Then it is straightforward to show that A,,
and A2X are distinct non trivial series classes of M(G) each containing a vertex
2-cocircuit of M(G). Thus by induction the required result is established. □
If one strengthens the hypothesis of the preceding result by insisting that the
graph G be minimally 2-connected, then one obtains a result of Dirac [5, Theorem
5] and Plummer [16, Corollary 2a]. The next three results follow from Theorem 2.9
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
52 J. G. OXLEY
in the same way that Corollaries 2.6 and 2.7 follow from Theorem 2.4. For a graph
G, the number of connected components of G will be denoted by k(G).
(2.10) Corollary. Let G be a 2-connected loopless graph other than a cycle.
Suppose that A G E(G) such that for all a in A, G \ a is not 2-connected. Then either
A is the edge-set of a forest, or V(A) contains at least \A\ — \V(A)\ + k(A) + 1
pairwise nonadjacent vertices having degree two in G.
(2.11) Corollary. Let G be a minimally 2-connected graph other than a cycle.
Then there is a set U containing at least \E(G)\ — \V(G)\ + 2 vertices such that each
member of U has degree two and every path in G joining two members of U contains a
vertex of degree greater than two.
(2.12) Corollary. A minimally 2-connected graph G having at least four edges has
at least \E(G)\ — |K(G)| + 2 pairwise nonadjacent vertices of degree two.
The last result can also be deduced from the results of Dirac [5, Theorem 5] and
Plummer [16, Corollary 2a] which Theorem 2.9 extends. We now compare the
bound in Corollary 2.12 with the following bound which comes from a result of
Dirac [5, (6), (5)], noting that the last result can also be used to give an alternative
derivation of Corollary 2.12.
(2.13) Theorem [5, (6), (5)]. A minimally 2-connected graph G has at least
(| V(G)\ + 4)/3 vertices of degree two.
The number of vertices of degree A: in a graph G will be denoted by vk.
(2.14) Proposition. Let G be a minimally 2-connected graph having at least four
edges. Then
v2 >
\E(G)\ - | V(G)\ + 2 for 4lK(S)|-2 < \E(G)\ < 2| V(G)\ - 4.
Proof. Dirac [5, (7)] has shown that E(G) contains at least | K(G)| and at most
2| V(G)\ - 4 edges. By Corollary 2.12 and Theorem 2.13,
v2 > max{(\V(G)\ + 4)/3, \E(G)\ - \V(G)\ + 2),
and it is straightforward to check that \E(G)\ - \V(G)\ + 2 > (\ V(G)\ + 4)/3 if
and only if \E(G)\ > (4\ V(G)\ - 2)/3. Thus
v2>
Í^f±* for\V(G)\<\E(G)\<W^,
\E(G)\ -1 V(G)\ + 2 for 4|F(G)I- < \E(G)\ < 2| K(G)| - 4.
But if \E(G)\ < (4| K(G)| - 2)/3, then
2\E(G)\< (8|F(G)|-4)/3. (2.15)
Now, by summing the vertex degrees of G we get
2\E(G)\>2v2 + 3(\V(G)\-v2). (2.16)
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
CONNECTIVITY IN MATROIDS AND GRAPHS 53
On combining (2.15) and (2.16), we get (8| K(G)| - 4)/3 > 3|F(G)| - v2. Hence
v2 > (| K(G)| + 4)/3, and therefore, v2 > (| V(G)\ + 5)/3. □
The graph K2J is one of infinitely many graphs which show that in Theorem 2.13
we cannot replace (\V(G)\ + 4)/3 by (|K(G)| + 5)/3 (see [1, pp. 15-16 and p. 48,
Exercise 20]), although clearly this replacement can be made unless | V(G)\ = 2
(mod 3). The structure of those minimally 2-connected graphs G having exactly
{(| V(G)\ + 4)/3} vertices of degree two will be determined elsewhere.
It is natural to ask whether Proposition 2.14 can be extended to minimally
«-connected graphs for « > 3. The proof of Corollary 2.12 relied heavily on the
fact that in a minimally 2-connected graph with at least four edges, each cycle
meets at least two nonadjacent vertices of degree two. The corresponding, but
slightly weaker, result for minimally «-connected graphs is an immediate conse-
quence of Theorem 2.8. Mader [8, Satz 2] used this result to show that in a
minimally «-connected graph G,
(n - l)\V(G)\ + 2" -2« - 1-' V"11)
Recently, Mader [9, Theorem 17] has strengthened (2.17) to get that
(«-l)[F(G)| + 2«
2« - 1 (2.18)
The following application of Theorem 2.8, although straightforward, seems new,
yet the bound it gives frequently sharpens (2.18). The proof is similar to the proof
of Corollary 2.6.
(2.19) Proposition. Let G be a minimally n-connected graph where « > 2. 77ie«
\E(G)\-\V(G)\ + l" -«^n-'
Proof. Let X be the set of edges of G which are incident with some vertex of
degree «. Then, by Theorem 2.8, X meets every cycle of G. That is, X meets every
circuit of M(G), and thus X contains a cobase B* of M(G). Now the set of edges
incident with a vertex u of G is a cocircuit of M(G), so B* contains at most « — 1
of the edges of a vertex «-cocircuit of G. Hence vn > \B*\/(n — I). As |5*| =
\E(G)\ - | V(G)\ + 1, the required result follows. \J
In the case « = 3, (2.18) was first proved by Halin [7, Satz 6]. His proof was
based on the fact that a cycle in a minimally 3-connected graph meets at least two
vertices of degree three [7, Satz 5]. Using this, the proof method of Propositions
2.14 and 2.19 can now be extended to give the following analogue of Proposition
2.14.
(2.20) Proposition. Let G be a minimally 3-connected graph. Then
v3 >
MQLil /orJHÇ>[<|£(c)|<^(g)l-3t
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
54 J. G. OXLEY
Proof. As every vertex of G has degree at least three, |£(G)| > 3|F(G)|/2.
Moreover, \E(G)\ < 3\V(G)\ - 6 (see, for example, [14, Theorem 4.7]). We shall
now show that i>3 > ¿(\E(G)\ - | V(G)\ + 3). The rest of the proof then follows by
generalizing the proof of Proposition 2.14. Let X and B* be as in the proof of
Proposition 2.19. Then we need to show that v3 > \(\B*\ + 2). Choose an element a
of B*. Then a is in a vertex 3-cocircuit C*. Now let Ca be the fundamental circuit
of a with respect to E(G)\ B*. Then Ca meets a vertex 3-cocircuit Cf which is
different from Cf. Since \C* n Ca\ > 2 for i = 1, 2, \B* n (Cf u C|)| < 3.
Therefore, as B* contains at most two elements of any vertex 3-cocircuit, v3
> \(\B* \ (Cx* u C2*)|) + 2. That is, t>3 > \(\B*\ + 1) with equality holding only if
\B* n C,*| = 2 and \B* n (C* U C*)\ = 3. Thus we may assume that there is an
element A in (B* n C*) \ Cf, for otherwise the required result holds. Now let Cb be
the fundamental circuit of A with respect to E(G) \ B*. Then Cb meets a vertex
3-cocircuit C* different from C*. Moreover, since A G Cx, it follows that C* ¥= C*.
Now \B* \ (Cf U C* u C*)\ > \B*\ - 4, so v3 > \(\B*\ - 4) + 3; that is, v3
> \(\B*\ + 2), as required. □
A natural question now is what can be said when « > 3 about the occurrence of
«-cocircuits in minimally «-connected matroids. Some progress has been made in
this direction for « = 3 and « = 4 [14], [25].
(2.21) Theorem [14, Theorem 2.5]. // C is a circuit of a minimally 3-connected
matroid M and \E(M)\ > 4, then M has at least two distinct 3-cocircuits meeting C.
It is not difficult to find examples showing that this result is best-possible. Using
it and the proof technique of Proposition 2.20, one can deduce that a minimally
3-connected matroid M with at least four elements has at least ^cork M + 1
3-cocircuits. This result is also best-possible. Some partial results on the question of
whether a minimally 4-connected matroid has a 4-cocircuit have been obtained by
Wong [25].
The results earlier in this section generalize known results for minimally 2-con-
nected matroids and graphs by weakening the requirement of minimal 2-con-
nectedness. In this context one is led to ask whether the hypothesis of Theorem
2.21 can be weakened so that one does not require the matroid M to be minimally
3-connected, but only that the deletion of elements of the circuit C should destroy
3-connectedness. The corresponding problem for graphs, which if true would
strengthen Theorem 2.8 in the case « = 3, also seems to be open.
3. A characterization of minimally 2-connected matroids. The main result of this
section solves a problem of Murty [10, p. 53] by providing a rather elementary yet
useful characterization of minimally 2-connected matroids. This characterization is
then used to give short proofs of the main results of Murty's paper [10]. We shall
use the operation of series connection of matroids and a number of properties of
this operation proved by Brylawski [3]. Let A/, and M2 he matroids having disjoint
ground sets Sx and S2 respectively. Choose an element/7, from Sx and an element
p2 from 52. These elements are called the basepoints of M, and M2 respectively. Let
p be a new element. Then the series connection S((MX; px), (M2; pj)) of M, and M2
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
CONNECTIVITY IN MATROIDS AND GRAPHS 55
with respect to the basepoint px and p2 is the matroid on (Sx \px) u (S2 \pj) u p
whose circuits are the circuits of Mx not containing px, the circuits of M2 not
containing p2, and all sets of the form (C, \pj) U (C2 \pj) u p where px G C,,
p2 G C2 and C, and C2 are circuits of Mx and M2 respectively. If the basepoints are
clear the series connection will sometimes be denoted S(MX, Mj). Series connection
and the corresponding dual operation, parallel connection, were studied in detail
by Brylawski [3]. More recently, Seymour [19] has looked at the closely related
operation of 2-sum of matroids, the 2-sum of Mx and M2 being precisely
S(MX, Mj)/p [22].
(3.1) Theorem. A matroid M is minimally 2-connected if and only if \E(M)\ > 3,
and either M is 2-connected and every element of M is in a 2-cocircuit, or
M = S((Mx/qx; px), (M2/q2; pj)) where both Mx and M2 are minimally 2-connected
having at least four elements and {px, qx) and {p2, q2) are cocircuits of Mx and M2
respectively.
Proof. Evidently if M is a 2-connected matroid for which \E(M)\ > 3 and every
element is in a 2-cocircuit, then M is minimally 2-connected. Now if M, and M2
are minimally 2-connected matroids each having at least three elements, then, by
Theorem 2.4, each certainly possesses a 2-cocircuit. Consider M = S((Mx/qx; pj),
(M2/q2; pj)) where {/>,, qx) and {p2, q2) are cocircuits of Mx and M2 respectively.
Then since, by Lemma 2.1, Mx/qx and M2/q2 are 2-connected, it follows by [3,
Proposition 4.6] that M is 2-connected.
We now show that M is minimally 2-connected. Firstly, M \p = (Mx/ qx) \px ®
(M2/q2)\p2 [3, Proposition 4.9], so M \p is not 2-connected. Now suppose that
e G E(Mx/qx) \px and consider M \ e. We have M \ e = S(Mx/qx, M2j'qj) \ e =
S(Mx/qx \ e, M2/q2) where the last step holds by [3, Proposition 4.7]. If Mx/qx \ e
is not 2-connected, then by [3, Proposition 4.6], M \ e is not 2-connected, as
required. Thus suppose that Mx/qx\ e is 2-connected. Then by Lemma 2.2, qx is a
loop or a coloop of Mx \ e. But M, has no loops or coloops, so {e, qx) is a cocircuit
of Mx. Therefore, as {px, qx) is a cocircuit, {e,px) is also a cocircuit of M, and
hence is a cocircuit of Mx/qx. Thus/?, is a coloop of Mx/qx \ e and thus, by [3,
Proposition 4.5], M \ e is not 2-connected. It follows that M is minimally 2-con-
nected, as required.
For the converse, suppose that M has an element p which is not in a 2-cocircuit.
Now M \ p is not 2-connected and since M has no 2-cocircuits containing p, the
matroid M \p has no component with fewer than two elements. Let M \p = TV,
© N2. Then, by [3, Proposition 4.10], M = S(M/E(Nj), M/E(N2)) where/? is the
basepoint of both M/E(NX) and M/E(N2). Let N3 = M/E(NX) and N4 =
M/ E(N2). Then each of N3 and N4 has at least three elements. Now, as M =
S(N3, A^4) and M is 2-connected, by [3, Proposition 4.6] again, each of N3 and 7V4 is
also 2-connected. If e G E(M) and e =^p, then e G E(N3) \p or e G E(N4) \p, so
suppose the former. Then as M \ e = S(N3 \ e, Ar4) and M \ e is not 2-connected,
N3\ e is not 2-connected. Thus for all elements e of #3, except possibly p, the
matroid N3\ e is not 2-connected. Similarly, N4\f is not 2-connected for all
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
56 J. G. OXLEY
elements/of N4 except possibly/?. Now for / = 1, 2, add an element q¡ in series
with/? in Ni+2 to get a new matroid M¡ which is clearly 2-connected. In fact, it is
not difficult to check that M, is minimally 2-connected. Then M can be obtained
by contracting qx and q2 from Mx and M2 respectively and then taking the series
connection of Mx/qx and M2/q2 with respect to the basepoint/?. Finally, as each
of N3 and N4 has at least three elements, each of Mx and M2 has at least four
elements. □
(3.2) Corollary [10, Theorems 3.2 and 3.4]. If r > 3, a minimally 2-connected
matroid M of rank r has at most 2r — 2 elements, the upper bound being attained if
and only if M = M(K2r^ j).
Proof. We argue by induction on \E(M)\. If every element of M is in a
2-cocircuit, then consider M*. Deleting a single element from every parallel class of
M* leaves a 2-connected matroid N having the same rank as M*. Evidently N has
at least rk M* + 1 elements with equality being attained only if A7 is a circuit. Thus
\E(M*)\ = \E(M)\ > 2(rk M* + 1) = 2(\E(M)\ - r + 1), and therefore \E(M)\ <2(r — 1) with equality being attained only if M* s Cj_x, an (r — l)-circuit in
which each element has been replaced by a pair of parallel elements. But if
M* s Cr2_„ then M s M(K2r_x).
We may now suppose that M has an element p which is not in a 2-cocircuit.
Then, by Theorem 3.1, M = S((Mx/qx; pj), (M2/q2; pj)) where, for / = 1, 2,
{/?,, q¡) is a cocircuit of M¡, and M, is minimally 2-connected having at least four
elements and hence having rank at least three. Now, by [3, Theorem 6.16(i)],
rk M = rk(Mx/qx) + rk(M2/q2) and thus
rk M = rk M, + rk M2 - 2. (3.3)
Moreover,
|£(M)|=|£(A/,)|+|£(M2)|-3, (3.4)
and, by the induction assumption, \E(M¡)\ < 2 rk A/,-— 2 for /' = 1,2. Thus
\E(MX)\ + \E(M2)\ < 2(rk A/, + rk M2 - 2), and, by (3.3) and (3.4), \E(M)\ < 2r
— 3. Hence, by induction, the required result is proved. □
Theorem 3.1 may also be used to give alternative proofs of several results for
minimally 2-connected matroids such as Corollary 2.7.
4. Colouring. The chromatic number of a loopless graph is the least positive
integer at which the value of its chromatic polynomial is positive. For loopless
matroids in general there are some difficulties in defining the chromatic number
(see [23, p. 264]). However, such problems do not arise for regular matroids. Thus if
M is a loopless regular matroid having chromatic polynomial P(M; X) (see, for
example, [23, p. 262]) its chromatic number x(^0 is nrin{y G Z+: P(M;j) > 0). It
can be shown (see, for example, [11, p. 17]) that, as for graphs, P(M; k) > 0 for all
integers k such that k > x(^0-
For a loopless matroid M representable over a finite field GF(q), Crapo and
Rota [4, Chapter 16] introduced an important invariant which, when M is regular,
is closely related to its chromatic number. The critical exponent c(M; q) of M is
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
CONNECTIVITY IN MATROIDS AND GRAPHS 57
min{y G Z+: P(M; q>) > 0}. It follows from [4, p. 16.4] that P(M; qk) > 0 for all
positive integers k and moreover, P(M; qk) > 0 if k > c(M; q).
The main result of this section is the following.
(4.1) Theorem. Let M be a minimally n-connected matroid where « = 2 or 3. If M
is representable over GF(q), then c(M; q) < 2 for q < « and c(M; q) = 1 for q > n.
Moreover, if M is regular, then x(^f ) < " + 1-
The proof of this will use three lemmas.
(4.2) Lemma. Let M be a minimally n-connected matroid where n = 2 or 3 and
suppose T G E(M). Then M\T has a cocircuit having at most n elements.
Proof. If T does not contain a circuit of M, then M \ T is free, so M \ T has a
coloop. If T does contain a circuit C of M, then, by Theorems 2.4 and 2.21, C
meets an «-cocircuit C* of M. Now C* certainly contains a cocircuit of M\T and
the required result follows. □
(4.3) Lemma [12, Lemma 5]. If M is a matroid representable over GF(q) and M is
minimal having critical exponent k + 1, then every cocircuit of M has at least qk
elements.
The analogue of the preceding result for regular matroids is as follows.
(4.4) Lemma [13, Theorem 3]. // M is a regular matroid which is minimal having
chromatic number k + I, then every cocircuit of M has at least k elements.
Proof of Theorem 4.1. We shall prove the result for Af representable over
GF(q) by using Lemmas 4.2 and 4.3. The result for M regular follows similarly by
using Lemma 4.4 in place of Lemma 4.3.
Suppose c(M; q) = k + 1. Then by deleting elements from M, we obtain a
restriction N which is minimal having critical exponent k + 1. By Lemma 4.3,
every cocircuit of N has at least qk elements. But, by Lemma 4.2, A^ has a cocircuit
having at most « elements. Thus n > qk and the required result follows. □
References
1. B. Bollobas, Extremal graph theory, London Math. Soc. Mono., no. 11, Academic Press, London,
New York and San Francisco, 1978.
2. J. A. Bondy and U. S. R. Murty, Graph theory with applications, Macmillan, London; American
Elsevier, New York, 1976.
3. T. H. Brylawski, A combinatorial model for series-parallel networks. Trans. Amer. Math. Soc. 154
(1971), 1-22.4. H. H. Crapo and G.-C. Rota, On the foundations of combinatorial theory: combinatorial geometries,
Preliminary ed., MIT Press, Cambridge, Mass, and London, 1970.
5. G. A. Dirac, Minimally 2-connected graphs, J. Reine Angew. Math. 228 (1967), 204-216.
6. J. E. Graver and M. E. Watkins, Combinatorics with emphasis on the theory of graphs, Springer-
Verlag, New York, Heidelberg and Berlin, 1977.
7. R. Haiin, Untersuchungen über minimale n-fach zusammenhangende Graphen, Math. Ann. 182
(1969), 175-188.8. W. Mader, Ecken vom Grad n in minimalen n-fach zusammenhangenden Graphen, Arch. Math.
(Basel) 23 (1972), 219-224.
9._, Connectivity and edge-connectivity infinite graphs. Surveys in Combinatorics (B. Bollobas,
ed.), Cambridge Univ. Press, New York, 1979, pp. 66-95.
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
58 J. G. OXLEY
10. U. S. R. Murty, Extremal critically connected matroids. Discrete Math. 8 (1974), 49-58.
11. J. G. Oxley, Colouring, packing and the critical problem, Quart. J. Math. Oxford Ser. (2) 29 (1978),
11-22.
12. _, On a covering problem of Mullin and Stanton for binary matroids, Aequationes Math. 20
(1980), 104-112.13. _, A generalization of a covering problem of Mullin and Stanton for matroids, Proc. Sixth
Austral. Combinatorics Conf., Lecture Notes in Math., vol. 748, Springer-Verlag, Berlin, Heidelberg and
New York, 1979, pp. 92-97.
14._, On matroid connectivity, Quart. J. Math. Oxford Ser. (2) (to appear).
15._, On ^-connected matroids, Cañad. J. Math, (to appear).
16. M. D. Plummer, On minimal blocks, Trans. Amer. Math. Soc. 134 (1968), 85-94.
17. P. D. Seymour, Matroid representation over GF(3), J. Combin. Theory. Ser. B 26 (1979), 159-173.
18. _, Packing and covering with matroid circuits, J. Combin. Theory Ser. B 28 (1980), 237-242.
19._, Decomposition of regular matroids, J. Combin. Theory Ser. B 28 (1980), 305-359.
20. W. T. Tutte, Connectivity in matroids, Cañad. J. Math. 18 (1966), 1301-1324.21. _, Wheels and whirls, Lecture Notes in Math., vol. 211, Springer-Verlag, Berlin, Heidelberg
and New York, 1971, pp. 1-4.
22. P. N. Walton and D. J. A. Welsh, On the chromatic number of binary matroids, Mathematika 27
(1980), 1-9.23. D. J. A. Welsh, Matroid theory, London Math. Soc. Mono., no. 8, Academic Press, London, New
York and San Francisco, 1976.
24. N. L. White, The bracket ring of a combinatorial geometry. II: Unimodular geometries, Trans.
Amer. Math. Soc. 214 (1975), 233-248.
25. P.-K. Wong, On certain n-connected matroids, J. Reine Angew. Math. 299/300 (1978), 1-6.
Department of Mathematics, IAS, Australian National University, Canberra, Australia
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use